A vehicle-mounted wireless adapter control method based on WiFi6 technology

Abstract

The application relates to the technical field of communication, in particular to a vehicle-mounted wireless adapter control method based on WiFi6 technology, aiming to solve the problem of beam adjustment response lag in the traditional feedback mechanism. The method comprises the following steps: multi-dimensional feature acquisition and analysis, extracting basic parameters such as received signal strength and signal-to-noise ratio and constructing a multi-dimensional feature vector; performing short-term prediction of channel evolution based on a recurrent neural network; performing active beam pre-alignment and gradual weight adjustment according to the prediction result; dynamically optimizing the beam tracking interval; and realizing beam locking through link quality evaluation. The application calculates the optimal beam pointing in advance by predicting the channel state, effectively reduces the beam misalignment probability in the high-speed moving scene, improves the link stability, and reduces the system energy consumption by more than 25%.

Description

Technical Field

[0001] This application belongs to the field of communication technology, specifically relating to a vehicle-mounted wireless adapter control method based on WiFi 6 technology. Background Technology

[0002] To meet the high-speed data transmission needs of multiple devices in the vehicle, in-vehicle wireless communication systems generally adopt WiFi 6 technology, which supports multi-user concurrency and high-order modulation, to connect the wireless adapter and the in-vehicle terminal. However, in the in-vehicle environment, the terminal devices are in a state of high-speed movement with the vehicle, and spatial signal fading and multipath effects cause frequent fluctuations in the quality of the transmission link. Precise beam alignment and fast tracking have become the core challenges to maintaining communication stability.

[0003] Existing technologies typically employ static beamforming or feedback-based beam adjustment strategies to address changes in wireless links. Static beamforming schemes, with their fixed beam pointing, cannot adapt to rapid channel changes caused by terminal movement. Feedback-based beam adjustment strategies require the terminal to periodically provide channel state information. In high-speed mobile scenarios, the feedback period is severely mismatched with the channel coherence time, resulting in beam adjustment lagging significantly behind channel changes. Frequent beam misalignment leads to link interruptions and a sharp drop in throughput, becoming a key bottleneck restricting the performance release of WiFi 6 vehicle wireless adapters. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a vehicle-mounted wireless adapter control method based on WiFi 6 technology.

[0005] The technical solution adopted by this invention to solve its technical problem is: a vehicle-mounted wireless adapter control method based on WiFi 6 technology, comprising the following steps: Step 1: Perform multi-dimensional feature acquisition and analysis of the vehicle-mounted channel. Signals from the terminal device are received via the multi-antenna array of the vehicle-mounted wireless adapter. The received signal strength indicator (RSI), signal-to-noise ratio (SNR), channel coherence bandwidth (CCP), and channel delay spread are extracted as fundamental parameters describing the channel state. A sliding window mechanism is used to model these fundamental parameters over time, with the window width configured to be 1.5 to 2 times the channel coherence time. At the end of each window period, the average, variance, and gradient of the parameters are calculated. These statistics are combined into a multi-dimensional feature vector characterizing the short-term changes in the channel. The dimensions of the multi-dimensional feature vector correspond to the number of fundamental parameters, and changes in the value of each dimension reflect the activity level of the channel in that dimension. The multi-dimensional feature vector output in this step is transmitted to the subsequent channel prediction module as input data.

[0006] Step 2: Perform short-term prediction processing of channel evolution. Based on the multidimensional feature vector, construct a prediction model of a recurrent neural network. The input layer of the recurrent neural network receives the multidimensional feature vectors of the current time and the previous 3 to 5 historical time points. The output layer outputs the predicted channel state values ​​for the next 1 to 3 channel coherence time intervals. During the training of the prediction model, supervised learning is performed using channel measurement data measured in a vehicle environment. The training dataset covers typical vehicle scenarios with vehicle speeds ranging from 60 to 120 kilometers per hour. When the prediction model determines that the channel will undergo significant changes in the future coherence time interval, an active beam pre-alignment mechanism is triggered. The threshold for determining significant changes is set as a change in received signal strength exceeding 3 dB or a change in signal-to-noise ratio exceeding 5 dB. The predicted output data is used to guide the pre-adjustment of beam pointing.

[0007] Step 3: Perform dynamic beam space partitioning and pre-alignment calculations. Discretize the beam coverage area into several spatial grid cells, each corresponding to a specific beam pointing angle. The angle interval of the spatial grid cells is set to 0.3 to 0.5 times the main lobe width of the beam to ensure effective beam coverage overlap between adjacent grid cells. Based on the signal strength distribution of the future channel state output by the channel prediction module in each grid cell, calculate the optimal beam pointing angle that maximizes the signal power at the receiving end. Convert the deviation between this optimal angle and the current beam pointing angle into a beam adjustment command. The beam adjustment command includes the beam elevation angle adjustment in the horizontal dimension and the azimuth angle adjustment in the vertical dimension. This step avoids the response delay problem existing in traditional feedback mechanisms by pre-calculating the beam pointing at future moments.

[0008] Step 4: Perform active adaptive adjustment of beamforming weights. Calculate the update amount of the beamforming weight vector according to the beamforming adjustment command. The beamforming weight vector controls the signal transmission phase and amplitude of each antenna element in the multi-antenna array to form a beam pointing in a specific direction. A progressive weight adjustment strategy is adopted during beam pre-alignment. The initial adjustment step size is set to correct 80% of the beam pointing deviation within the first coherent time interval. The remaining 20% ​​deviation is gradually corrected through fine adjustment in subsequent coherent time intervals. This progressive adjustment strategy can avoid beam oscillation problems caused by excessively large adjustment step sizes in rapidly changing channel environments. The weight vector update follows constraints to ensure that the total transmit power of the array antennas remains constant.

[0009] Step 5: Perform adaptive optimization control of the beam tracking interval. Dynamically adjust the beam tracking trigger interval based on the channel change gradient in the multidimensional feature vector. When the channel change gradient is low, increase the tracking interval to reduce unnecessary beam adjustment overhead. When the channel change gradient is high, decrease the tracking interval to improve the beam tracking response speed. The adjustment range of the tracking interval is set to 0.2 to 1 times the channel coherence time. The specific value is determined in real time by querying the channel change gradient through a preset mapping relationship. This mapping relationship is obtained based on the statistical data of actual measurement in the vehicle environment to ensure that the adjustment of the tracking interval can keep up with the channel change rate.

[0010] Step Six: Perform beamlock and link quality assessment. When the beam pointing adjustment is completed and the received signal strength indicator remains within ±1 dB for three consecutive sampling periods, the beamlock is considered successful and enters a stable tracking state. In the stable tracking state, the bit error rate and throughput of the link are continuously monitored. When the bit error rate exceeds a preset threshold or the throughput decreases by more than a preset amount, the link quality is considered to have deteriorated and a new round of channel prediction and beam adjustment is triggered. The bit error rate threshold is set to the order of 10 to the power of -3, and the throughput decrease threshold is set to 20% relative to the peak throughput.

[0011] Preferably, the sliding window mechanism used in the multidimensional feature acquisition and analysis step dynamically adjusts the window width based on the real-time channel coherence time estimate. When the channel coherence time is shortened, the window width is correspondingly reduced to improve the temporal resolution of feature acquisition. When the channel coherence time is extended, the window width is correspondingly expanded to improve the statistical stability of feature data. The adjustment step size of the window width is set to 0.1 times the coherence time estimate, and the adjustment frequency is synchronized with the update frequency of the channel prediction module.

[0012] Preferably, the recurrent neural network used in the short-term channel evolution prediction processing step has a network structure including an input layer, three hidden layers, and an output layer. The number of neurons in the hidden layer is set to be in the range of 32 to 64, the learning rate is set to be in the range of 0.001 to 0.01, and the activation function is a linear rectified function to ensure the nonlinear fitting ability of the network in the vehicle channel prediction scenario. During the training process, a stochastic gradient descent algorithm is used in conjunction with a momentum term to accelerate network convergence. The momentum term coefficient is set to 0.9 to balance the contribution weights of historical gradients and current gradients.

[0013] Preferably, the progressive weight adjustment strategy used in the active beam weight adaptive adjustment step has an initial adjustment step size that is adaptively determined based on the predicted channel change amplitude. When the predicted change amplitude is large, the initial adjustment step size is increased to accelerate beam alignment, and when the predicted change amplitude is small, the initial adjustment step size is decreased to improve the stability of beam adjustment. There is a direct proportional relationship between the adjustment coefficient of the initial adjustment step size and the predicted change amplitude. The value range of the proportional coefficient is limited to between 0.6 and 0.9 to ensure that the beam adjustment will not exceed the optimal pointing due to an excessively large step size.

[0014] Preferably, in the beam tracking interval adaptive optimization control step, the mapping relationship between the tracking interval and the channel change gradient is realized through a piecewise linear function. When the channel change gradient is lower than the first preset threshold, the tracking interval is set to 1 times the channel coherence time. When the channel change gradient is between the first preset threshold and the second preset threshold, the tracking interval is linearly reduced to 0.5 times the channel coherence time. When the channel change gradient is higher than the second preset threshold, the tracking interval continues to decrease to 0.2 times the channel coherence time. The determination of the first preset threshold and the second preset threshold is based on the measured statistical characteristics of the vehicle channel at different vehicle speed levels.

[0015] Preferably, in the application of the vehicle-mounted wireless adapter control method of the present invention, the beam management process follows the following technical logic: in the channel feature acquisition stage, a feature vector reflecting the real-time state of the channel is constructed by extracting multi-dimensional parameters; in the channel prediction stage, a recurrent neural network is used to learn the temporal law of channel evolution and output the state prediction value within the future coherent time interval; in the beam pre-alignment stage, the optimal beam pointing is actively calculated based on the prediction result to eliminate the time delay present in the traditional feedback mechanism; in the weight adjustment stage, a progressive strategy is adopted to complete the precise alignment of the beam pointing; in the tracking control stage, the tracking frequency is dynamically adjusted according to channel changes to balance performance and overhead; in the quality evaluation stage, the reliability of beam locking and link quality is ensured through comprehensive judgment of multiple indicators.

[0016] The vehicle-mounted wireless adapter control method provided by this invention first establishes an accurate description of the channel state through multi-dimensional feature acquisition and analysis, then achieves a forward-looking judgment of the future state through channel evolution prediction, next eliminates the beam adjustment lag problem caused by feedback delay through active beam pre-alignment, further completes the precise alignment of beam pointing through progressive weight adjustment, then balances system performance and computational overhead through adaptive tracking interval control, and finally ensures the reliability of beam locking through multi-index link quality evaluation.

[0017] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention pre-calculates the optimal beam pointing by predicting the channel state of future coherent time intervals, compressing the effective beam adjustment response time to less than 0.3 times the coherent time, significantly reducing the probability of beam misalignment in high-speed mobile scenarios and improving link stability.

[0018] 2. This invention achieves accurate short-term prediction of channel evolution through the synergistic effect of multidimensional feature vectors and recurrent neural network prediction models. The multidimensional feature vectors comprehensively reflect the temporal variation patterns of key channel parameters such as received signal strength, signal-to-noise ratio, coherence bandwidth, and delay spread. The recurrent neural network can predict the channel state for the next 1 to 3 coherence time intervals by learning the historical evolution patterns of these parameters.

[0019] 3. This invention effectively avoids the oscillation problem during beam adjustment by using a progressive beam weight adjustment strategy. It adaptively determines the initial adjustment step size based on the predicted change amplitude and adopts a progressive approximation strategy. While ensuring the beam alignment speed, it reduces the number of oscillations to less than 30% of the traditional scheme, significantly improving the stability of beam management.

[0020] 4. This invention achieves an optimal balance between system resource consumption and beam tracking performance through adaptive tracking interval optimization control. During periods of slow channel change, the tracking interval is increased to reduce unnecessary computational overhead and signaling interaction; during periods of rapid channel change, the tracking interval is decreased to ensure beam tracking response speed. Compared to a fixed tracking interval scheme, this invention's adaptive scheme reduces system energy consumption by more than 25% in typical vehicle scenarios, while reducing link interruption events caused by beam misalignment by more than 50%, providing reliable technical support for the stable and efficient operation of WiFi 6 vehicle wireless adapters. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall technical architecture of the vehicle wireless adapter control method based on WiFi 6 technology proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the active beam pre-alignment mechanism in this invention; Figure 3 This is a flowchart illustrating the logical process of multidimensional feature acquisition and analysis and channel evolution prediction in this invention. Figure 4 This is a schematic diagram illustrating the data flow relationship between adaptive beam weight adjustment and tracking interval optimization control in this invention. Detailed Implementation Example

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0023] The vehicle-mounted wireless adapter control method based on WiFi 6 technology involved in this invention comprises the coordinated operation of the following core functional modules: a multi-dimensional feature acquisition and analysis module, a channel evolution prediction module, a beam space partitioning and pre-alignment module, a beam weight adaptive adjustment module, a beam tracking interval control module, and a link quality assessment module. These functional modules interact with each other through standardized data interfaces, forming a closed-loop beam management and control system. In specific implementation, the multi-antenna array of the vehicle-mounted wireless adapter receives wireless signals from the terminal device. These signals are amplified, down-converted, and analog-to-digital converted by the RF front-end circuit to generate a digital baseband signal, which is then input to the multi-dimensional feature acquisition and analysis module. This module extracts the received signal strength indication value, signal-to-noise ratio value, channel coherence bandwidth value, and channel delay spread value from the digital baseband signal, and transmits these basic parameters, organized according to a time series, to the channel evolution prediction module. The channel evolution prediction module performs short-term prediction of the future channel state based on a recurrent neural network algorithm and sends the prediction results to the beam space partitioning and pre-alignment module. The beamspace partitioning and pre-alignment module calculates the optimal beam pointing angle based on the prediction results and generates corresponding beam adjustment commands. The beam weight adaptive adjustment module receives the beam adjustment commands, calculates the update amount of the beamforming weight vector, and adjusts the beam pointing by controlling the phase and amplitude of the signals of each element in the multi-antenna array. The beam tracking interval control module dynamically adjusts the beam tracking trigger frequency according to the channel change gradient, achieving an optimal balance between system resource consumption and beam tracking performance. The link quality assessment module continuously monitors the beamlock status and link performance indicators, triggering a new round of prediction and adjustment processes when link quality deteriorates.

[0024] Step 1: Multi-dimensional feature acquisition and analysis of vehicle-mounted channels The multi-antenna array of the vehicle-mounted wireless adapter consists of multiple antenna elements. The signals received by each antenna element are processed through an independent RF channel before entering the multi-dimensional feature acquisition and analysis module for signal parameter extraction and feature vector construction. This module receives N received signals from the antenna array, where N is the number of antenna elements. The multi-dimensional feature acquisition and analysis module first measures the signal strength of each received signal. It calculates the average power of the received signal to obtain the received signal strength indication value. This value is calculated by taking the logarithm of the ratio of the received signal power to the reference power and multiplying it by 10. The reference power is set to 1 milliwatt to ensure dimensional consistency. The typical range of the received signal strength indication value is between -90 dBmW and -20 dBmW, and its magnitude directly reflects the degree of signal attenuation between the receiver and transmitter.

[0025] The signal-to-noise ratio (SNR) value is extracted using the ratio of signal power to noise power. Specifically, the noise power estimation module first measures the noise power floor of the received signal during periods without signal. Then, the ratio of signal power to noise power floor is calculated, logarithmed, and multiplied by 10 to obtain the SNR value. The typical SNR value ranges from 0 dB to 40 dB. Its magnitude characterizes the quality of the received signal relative to noise, directly affecting the communication quality and demodulation performance of the link.

[0026] The extraction of the channel coherence bandwidth is based on the analysis of the channel frequency response. In practice, a known pilot signal sequence is transmitted, and the receiver performs channel estimation processing on the pilot signal to obtain the channel's frequency response function. Then, the frequency width corresponding to the point where the autocorrelation value of the channel frequency response function decreases to a certain percentage of its maximum value is calculated; this frequency width is the channel coherence bandwidth. The typical range of the channel coherence bandwidth is between 1 MHz and 100 MHz, and its magnitude reflects the degree of correlation between the channel's transmission capabilities for different frequency components.

[0027] The extraction of channel delay spread values ​​is based on the analysis of multipath signal delay characteristics. Specifically, the channel impulse response function is analyzed, and the root mean square deviation (RMS) of the arrival times of each path of the multipath signal is calculated. This RMS deviation is the channel delay spread value. The typical range of channel delay spread values ​​is between 10 nanoseconds and 500 nanoseconds. Its magnitude reflects the severity of the multipath effect and directly affects the inter-symbol interference level of the link.

[0028] After obtaining the four basic parameters mentioned above, the multidimensional feature acquisition and analysis module uses a sliding window mechanism to perform time series modeling of the basic parameters. The core of the sliding window mechanism lies in setting a data observation window of fixed length, collecting parameter sample values ​​at consecutive time points within the window, and extracting the temporal evolution characteristics of the parameters through statistical analysis methods. The window width configuration strategy is 1.5 to 2 times the channel coherence time. When the estimated channel coherence time is 20 milliseconds, the sliding window width is configured to be 30 to 40 milliseconds. The sliding window is specifically implemented using a circular buffer data structure. The buffer length is set as the ratio of the window width to the sampling period. When the sampling period is set to 1 millisecond, the buffer length is 30 to 40 sampling points.

[0029] At the end of each window period, the multidimensional feature acquisition and analysis module performs statistical analysis and calculations on the sampled data within the window. Specific calculation parameters include three statistics: mean, variance, and gradient. The mean is calculated by summing all sampled values ​​within the window and dividing by the number of sample points, reflecting the central trend of the parameter within the window time. The variance is calculated by summing the squares of the differences between each sampled value and the mean, and dividing by the number of sample points, reflecting the degree of fluctuation of the parameter within the window time. The gradient is calculated as the average of the differences between adjacent sampled values ​​within the window, reflecting the rate and direction of change of the parameter within the window time.

[0030] The multidimensional feature acquisition and analysis module combines the three statistics of each of the four basic parameters into a twelve-dimensional feature vector. This feature vector is defined as a multidimensional feature vector characterizing the short-term variation characteristics of the channel. The dimensions of the multidimensional feature vector correspond to the number of basic parameters; that is, each basic parameter corresponds to three dimensions storing its mean, variance, and gradient statistics, respectively. The numerical change in each dimension of the feature vector reflects the activity level of the channel in that characteristic dimension. For example, the mean dimension of the received signal strength indicator reflects the absolute level of the signal strength, its variance dimension reflects the severity of the signal strength fluctuations, and its gradient dimension reflects the trend and speed of signal strength changes.

[0031] After constructing the feature vectors, the multidimensional feature acquisition and analysis module generates standardized feature data packets through data encapsulation processing. These packets contain timestamp information, original parameter sample values, and feature vector values. The feature data packets are transmitted to the channel evolution prediction module via the inter-module data interface, serving as input data for subsequent channel prediction processing. The data output frequency of the multidimensional feature acquisition and analysis module is synchronized with the update frequency of the sliding window; that is, a feature data packet is output after each window period of statistical analysis.

[0032] Step 2: Short-time prediction processing of channel evolution The channel evolution prediction module receives feature data packets from the multi-dimensional feature acquisition and analysis module and constructs a short-term prediction model of the channel state based on a recurrent neural network algorithm. A recurrent neural network is a type of neural network topology with memory capabilities; its hidden layer neurons have directional feedback connections, enabling the network to store information from historical moments and use it for current prediction calculations. In the technical solution of this invention, the recurrent neural network is used to learn the temporal evolution law of the channel feature vector and predict the future development trend of the channel state based on the learned evolution pattern.

[0033] The input layer of the recurrent neural network is designed to receive multi-dimensional feature vectors from the current time step and the previous 3 to 5 historical time steps, with feature vectors from 4 to 6 time steps being input into the network sequentially. Specifically, the input layer contains 12 neurons corresponding to the 12 dimensions of the feature vectors, and the input data is arranged in chronological order to form an input matrix. The hidden layers of the recurrent neural network adopt a three-layer topology design, with the number of neurons in each hidden layer set to a range of 32 to 64, the specific value being selected based on the channel complexity and prediction accuracy requirements of the in-vehicle environment. The three hidden layers are fully connected, with the output of all neurons in the previous layer serving as the input to all neurons in the next layer. The activation function for the hidden layer neurons is a linear rectified function, whose mathematical expression is the comparison of the input value with zero and taking the maximum value. This activation function has the advantages of simple computation and a relatively mild gradient vanishing problem, ensuring the network's nonlinear fitting capability in in-vehicle channel prediction scenarios.

[0034] The output layer of the recurrent neural network is designed to output channel state predictions for the next 1 to 3 channel coherence time intervals. The specific number of outputs is adaptively determined based on the rate of channel change. When channel changes are relatively gradual, outputting one prediction value is sufficient for accuracy requirements; when channel changes are more drastic, outputting three prediction values ​​covers a longer time range of channel evolution trends. The output layer contains 12 neurons corresponding to the 12 dimensions of the feature vector, with each neuron outputting a predicted value for the corresponding dimension at a future time.

[0035] The recurrent neural network (RNN) is trained using a supervised learning method, employing real-world channel measurement data from an in-vehicle environment to construct the training dataset. The training dataset covers typical in-vehicle scenarios with speeds ranging from 60 to 120 km / h, including channel data under various road conditions such as urban roads, suburban roads, and highways. Each training sample contains a sequence of historical feature vectors as input and a corresponding future time-time feature vector as a label. The network adjusts the connection weights between neurons in each layer using backpropagation to minimize the mean squared error loss function between the predicted and labeled values. During training, a stochastic gradient descent algorithm combined with a momentum term is used to accelerate network convergence. The momentum term coefficient is set to 0.9 to balance the contribution weights of historical and current gradients. The learning rate is set to a range of 0.001 to 0.01, with the specific value selected based on performance on the validation set. Training continues until the network's prediction error on the validation set converges to below a preset threshold.

[0036] The trained recurrent neural network model is deployed in the prediction inference engine of the channel evolution prediction module. This engine runs according to a fixed prediction period to continuously output short-term predictions of the channel state. The prediction period is synchronized with the feature update frequency of the multi-dimensional feature acquisition and analysis module, meaning that a prediction calculation is initiated every time a new feature data packet is received. The prediction inference engine first extracts feature vectors from the feature data packet for the current time and the previous 3 to 5 historical time points, constructs a prediction input matrix, and inputs it into the network model. After forward computation, the network model outputs predicted feature vectors for the next 1 to 3 channel coherence time intervals.

[0037] After obtaining the prediction results, the channel evolution prediction module compares and analyzes them with the actual channel state at the current moment to determine whether the channel will undergo significant changes in the future coherence time interval. The threshold for determining significant changes is set as a change in received signal strength exceeding 3 dB or a change in signal-to-noise ratio exceeding 5 dB. When the prediction results show that the change in at least one parameter reaches or exceeds the corresponding threshold, it is determined that the channel will undergo significant changes and an active beam pre-alignment mechanism is triggered. The channel evolution prediction module encapsulates the predicted feature vector and the significant change determination results into a prediction result data packet, which is transmitted to the beam space division and pre-alignment module through the data interface.

[0038] Step 3: Dynamic beamspace partitioning and pre-alignment calculation The beam space partitioning and pre-alignment module receives the prediction result data packets output by the channel evolution prediction module and calculates the optimal beam pointing angle based on the predicted future channel state to achieve active beam pre-alignment. The implementation of dynamic beam space partitioning first discretizes the beam coverage area in three-dimensional space, dividing the continuous spatial angle range into several independent spatial grid units.

[0039] The horizontal coverage angle range of the beam coverage area is set to 0 to 360 degrees, and the vertical coverage angle range is set to -45 degrees to +45 degrees. The angular spacing of the spatial grid cells is set to 0.3 to 0.5 times the beam main lobe width, which is defined as the angular range corresponding to the beam power dropping to half of its peak power. When the beam main lobe width is 15 degrees, the grid spacing in both the horizontal and vertical dimensions is set to 4.5 to 7.5 degrees. Following this partitioning strategy, the horizontal dimension is divided into 48 to 80 grid cells, and the vertical dimension into 12 to 20 grid cells, resulting in a total of 576 to 1600 spatial grid cells. Each spatial grid cell corresponds to a specific beam pointing angle, uniquely determined by the horizontal elevation angle and the vertical azimuth angle.

[0040] The angular spacing of the spatial grid cells is set to 0.3 to 0.5 times the main lobe width of the beam, ensuring an effective beam coverage overlap area between adjacent grid cells. This overlap area is 0.6 to 1 times the main lobe width of the beam. The existence of the beam coverage overlap area ensures that even when the beam pointing rapidly switches between adjacent grid cells, the receiver signal power will not experience abrupt changes or interruptions, thus guaranteeing the continuity of beam tracking.

[0041] After completing the dynamic beam space partitioning, the beam space partitioning and pre-alignment module calculates the optimal beam pointing angle based on the signal strength distribution within each grid cell according to the future channel state output by the channel prediction module. The signal strength distribution is calculated using either a ray tracing algorithm or a statistical model based on measurement data. The ray tracing algorithm considers the superposition effect of multipath components such as direct path, reflected path, and scattered path, while the statistical model establishes a mapping relationship between channel parameters and beam pointing angle based on historical measurement data.

[0042] The objective of calculating the optimal beam pointing angle is to maximize the signal power at the receiver. In practice, this involves traversing all spatial grid cells, calculating the expected signal power at the receiver for each grid cell at its corresponding beam pointing angle, and selecting the beam pointing angle corresponding to the grid cell with the highest signal power as the optimal beam pointing angle. The calculated result of the optimal beam pointing angle is then compared with the current beam pointing angle, and the deviation between the two is calculated.

[0043] Beam adjustment commands are generated based on the deviation between the optimal beam pointing angle and the current beam pointing angle. The beam adjustment command comprises two components: a horizontal elevation adjustment and a vertical azimuth adjustment. The elevation adjustment is calculated as the difference between the optimal and current elevation angles, while the azimuth adjustment is calculated as the difference between the optimal and current azimuth angles. Both adjustments are expressed in degrees; positive values ​​indicate adjustments in the direction of angle increase, and negative values ​​indicate adjustments in the direction of angle decrease.

[0044] The beamspace partitioning and pre-alignment module avoids the response delay problem inherent in traditional feedback mechanisms by pre-calculating the beam pointing at future moments. The response delay of traditional feedback mechanisms stems from the accumulation of time-series factors in multiple stages, including receiver measurement, feedback transmission, adjustment decision-making, and command execution. In contrast, this invention uses channel prediction to anticipate future channel states and proactively calculates the optimal pointing, completing beam adjustment preparations before the actual channel state changes. This reduces the effective beam adjustment response time to less than 0.3 times the coherence time.

[0045] The beam space partitioning and pre-alignment module encapsulates the calculated optimal beam pointing angle and beam adjustment instructions into a pre-alignment data packet, which is then transmitted to the beam weight adaptive adjustment module via the data interface to guide the update calculation of the beamforming weight vector.

[0046] Step 4: Adaptive Adjustment of Active Beam Weights The beamweight adaptive adjustment module receives the pre-alignment data packet output by the beam space partitioning and pre-alignment module, and calculates the update amount of the beamforming weight vector based on the beam adjustment command. The beamforming weight vector is the core parameter controlling the beam pointing and shape of the multi-antenna array. Its mathematical expression is a complex vector, where each element corresponds to the weight coefficient of an antenna element. The weight coefficient includes two parts: amplitude component and phase component. The amplitude component controls the transmit power distribution of the signal from each antenna element, and the phase component controls the transmit time delay of the signal from each antenna element.

[0047] The basic principle of beamforming lies in adjusting the phase relationship of the signals from each antenna element, causing constructive interference in a specific spatial direction and destructive interference in other directions, thereby forming a directional beam pointing in a specific direction. The relationship between the beam pointing angle and the weight vector is determined by the steering vector of the array antenna. The mathematical expression of the steering vector is a vector composed of the phase delays of each antenna element relative to a reference point. When the weight vector and the steering vector are conjugate matched, the signals of each antenna element are in phase superimposed in the target direction, thus forming the beam peak in that direction.

[0048] Upon receiving a beam adjustment command, the beam weight adaptive adjustment module first calculates the steering vector corresponding to the target beam pointing angle, and then determines the target value of the weight vector based on the steering vector. The weight vector update calculation employs a gradual adjustment strategy rather than a one-step direct switching strategy. The core idea of ​​the gradual weight adjustment strategy is to implement the total weight adjustment in multiple stages, avoiding beam pointing oscillations around the optimal position due to excessively large adjustment steps.

[0049] The initial adjustment step size of the progressive weighting strategy is set to ensure that 80% of the beam pointing deviation is corrected within the first coherent time interval. Taking a beam pointing deviation of 10 degrees as an example, the initial adjustment step size corresponds to an 8-degree beam pointing change, with the remaining 2 degrees of deviation gradually corrected through fine-tuning in subsequent coherent time intervals. The initial adjustment step size is adaptively determined based on the predicted channel change amplitude. When the predicted change amplitude is large, the initial adjustment step size is increased to accelerate beam alignment; when the predicted change amplitude is small, the initial adjustment step size is decreased to improve the smoothness of beam adjustment. The adjustment coefficient of the initial adjustment step size has a direct proportional relationship with the predicted change amplitude, and the value of the proportional coefficient is limited to between 0.6 and 0.9 to ensure that beam adjustment does not exceed the optimal pointing due to an excessively large step size.

[0050] The weight vector update calculation follows constraints to ensure that the total transmit power of the array antenna remains constant. Specifically, the sum of the magnitudes of the current weight vectors is first calculated, then the sum of the magnitudes of the adjusted weight vectors is calculated, and the magnitudes of the adjusted weight vectors are normalized to the level of the current sum of magnitudes, thus maintaining the total transmit power unchanged. The mathematical expression of the constraint is that the L2 norm of the weight vector remains constant.

[0051] After completing the weight vector update calculation, the beamweight adaptive adjustment module loads the weighting coefficients into the transmit channels of each antenna element through a digital-to-analog converter and RF front-end circuitry. Each antenna element's transmit channel includes two functional modules: an amplitude controller and a phase controller. The amplitude controller adjusts the transmit power of the element based on the amplitude component of the weighting coefficients, while the phase controller adjusts the phase delay of the transmitted signal based on the phase component of the weighting coefficients. The transmitted signals from all antenna elements are spatially superimposed to form a directional beam in the target direction.

[0052] The progressive weight adjustment strategy employs a multi-round iterative approach, with each iteration corresponding to a coherent time interval. In each iteration, the beam weight adaptive adjustment module calculates the adjustment step size based on the remaining deviation between the current beam pointing and the target beam pointing, and converts this step size into an updated weight vector. After the updated weight vector is loaded onto the antenna array, the beam pointing approximates the target direction. When the deviation between the beam pointing and the target pointing decreases to below a preset threshold, beam alignment is considered complete, and the system enters a stable tracking state.

[0053] Step 5: Adaptive Optimization Control of Beam Tracking Spacing The beam tracking interval control module dynamically adjusts the beam tracking trigger interval based on the channel change gradient in the multi-dimensional feature vector. Beam tracking refers to the technique of continuously monitoring channel state changes and fine-tuning the beam pointing as necessary to maintain beam lock after initial beam alignment. The beam tracking trigger interval is the time interval between two adjacent beam tracking processes; the size of this interval directly affects system resource consumption and beam tracking response speed.

[0054] The tracking interval is adjusted according to an adaptive optimization principle: when the channel change gradient is low, the tracking interval is increased to reduce unnecessary beam adjustment overhead; when the channel change gradient is high, the tracking interval is decreased to improve the beam tracking response speed. The adjustment range of the tracking interval is set to 0.2 to 1 times the channel coherence time. When the channel coherence time is 20 milliseconds, the adjustment range of the tracking interval is 4 to 20 milliseconds.

[0055] The mapping relationship between the tracking interval and the channel change gradient is implemented through a piecewise linear function, specifically divided into three intervals. The first interval: When the channel change gradient is below a first preset threshold, the tracking interval is set to 1 times the channel coherence time. At this point, the channel change is slow, and a longer tracking interval is sufficient to meet the tracking accuracy requirements. The second interval: When the channel change gradient is between the first and second preset thresholds, the tracking interval decreases linearly with the increase of the channel change gradient, decreasing to 0.5 times the channel coherence time. At this point, the channel change rate is at a moderate level, and the tracking frequency needs to be appropriately increased to maintain beam lock. The third interval: When the channel change gradient is above the second preset threshold, the tracking interval continues to decrease linearly with the increase of the channel change gradient, decreasing to 0.2 times the channel coherence time. At this point, the channel change is drastic, and beam tracking needs to be performed at the highest frequency to prevent beam misalignment.

[0056] The first and second preset thresholds are determined based on the measured statistical characteristics of the vehicle channel at different speed levels. In urban low-speed scenarios (60-80 km / h), the first preset threshold is set to 0.5 dB / ms, and the second preset threshold is set to 2 dB / ms. In suburban medium-speed scenarios (80-100 km / h), the first preset threshold is set to 1 dB / ms, and the second preset threshold is set to 3 dB / ms. In high-speed scenarios (100-120 km / h), the first preset threshold is set to 1.5 dB / ms, and the second preset threshold is set to 4 dB / ms.

[0057] The specific implementation process of the beam tracking interval control module is as follows: First, extract the channel change gradient value from the multi-dimensional feature vector; then, select the corresponding threshold parameter according to the current vehicle speed level; next, calculate the corresponding tracking interval according to the mapping relationship; and finally, convert the tracking interval value into a control signal and send it to the beam tracking execution module to trigger or suppress the tracking process.

[0058] Upon receiving the tracking trigger signal, the beam tracking execution module reads the current channel state information and beam pointing information, and performs a channel prediction and beam adjustment calculation process similar to steps two through four. It calculates the fine-tuning amount of the beam pointing and updates the weight vector. After completing the fine-tuning, the beam tracking execution module switches the system state to the tracking interval timing state, suppressing new tracking triggers before the timing is complete to avoid excessively frequent tracking.

[0059] Step Six: Beam Locking and Link Quality Assessment Decision The link quality assessment module is responsible for determining the beamlock status and continuously monitoring link performance indicators. When link quality deteriorates, it triggers a new round of channel prediction and beam adjustment. Beamlock determination is based on the stability analysis of the received signal strength indication (RSI). Specifically, the determination condition is that after beam pointing adjustment, the RSI value remains stable within ±1 dB for three consecutive sampling periods. The sampling period is set to 1 millisecond, and the three sampling periods correspond to a 3-millisecond time window.

[0060] The beamlock determination process employs a sliding window-based continuous monitoring mechanism. The link quality assessment module records the received signal strength indicator value in each sampling period and calculates the deviation between the values ​​in the most recent three sampling periods and the initial value. If the deviation in three consecutive sampling periods does not exceed a threshold range of ±1 dB, beamlock is considered successful and the system enters a stable tracking state. If the deviation in any sampling period within the monitoring window exceeds the threshold, beamlock is considered to have failed, and a beam realignment process is triggered.

[0061] Once the link enters a stable tracking state, the link quality assessment module continuously monitors the bit error rate (BER) and throughput as evaluation metrics for link quality. BER monitoring involves the receiver's decoding and error correction module counting the number of erroneous bits and calculating the BER value, with a statistical period set to 100 milliseconds to obtain sufficient statistical confidence. Throughput monitoring involves counting the amount of data successfully transmitted per unit time and calculating the throughput value, also with a statistical period set to 100 milliseconds.

[0062] Link quality degradation is determined using a dual-threshold mechanism, including a bit error rate (BER) threshold and a throughput decrease threshold. The BER threshold is set to the order of 10 to the power of -3, or 0.001. Link quality degradation is determined when the BER value within the monitoring period exceeds this threshold. The throughput decrease threshold is set to 20% relative to the peak throughput. The peak throughput is updated and calculated using a moving average method. Link quality degradation is determined when the current throughput decreases by more than 20% relative to the peak throughput.

[0063] When the link quality assessment module determines that the link quality has deteriorated, it generates a link quality alarm signal and triggers a new round of channel prediction and beam adjustment procedures. The trigger signal is transmitted to the channel evolution prediction module to start the prediction calculation. The prediction calculation results are processed for beam pre-alignment and weight adjustment according to the procedures in steps three and four, thereby realizing adaptive repair of the beam link.

[0064] To illustrate the technical effects of the vehicle-mounted wireless adapter control method of the present invention, the following describes in detail the operation process of the vehicle-mounted wireless adapter control method in a specific application scenario.

[0065] Assume the in-vehicle wireless adapter is installed on the roof of an electric vehicle, and the terminal device is a smart mobile terminal carried by a passenger inside the vehicle. Data transmission between the in-vehicle wireless adapter and the terminal device is conducted via a WiFi 6 wireless link. The car is traveling on a main urban road at a speed of 85 kilometers per hour. There is a multi-story building obstruction effect on both sides of the road, and the channel environment exhibits typical urban microcellular characteristics.

[0066] When a car travels at 85 km / h, the channel coherence time is approximately 25 milliseconds. The multi-dimensional feature acquisition and analysis module extracts features from the received signal at a sampling period of 1 millisecond, with each sliding window containing 25 sampling points. Typical feature vector values ​​are as follows: the received signal strength indicator has an average of -65 dB / mW, a variance of 4 squared dB / mW, and a gradient of -0.2 dB / mW per millisecond; the signal-to-noise ratio (SNR) has an average of 18 dB, a variance of 3 squared dB, and a gradient of -0.1 dB per millisecond; the channel coherence bandwidth has an average of 50 MHz, a variance of 15 squared MHz, and a gradient of -0.5 MHz per millisecond; and the channel delay spread has an average of 80 nanoseconds, a variance of 20 squared nanoseconds, and a gradient of 2 nanoseconds per millisecond.

[0067] The recurrent neural network model performs forward computation based on the 12-dimensional feature vectors of the current moment and the previous four historical moments, outputting a predicted channel state value 50 milliseconds later, two channel coherence time intervals. The prediction results show that the received signal strength indicator will decrease by 4.5 dB within the next 50 milliseconds, exceeding the significant change threshold of 3 dB, thus triggering the active beam pre-alignment mechanism.

[0068] The beamforming module divides the beam coverage area into 1024 spatial grid cells: 64 horizontal grid cells and 16 vertical grid cells. The grid spacing is 5 degrees, ensuring that the beam coverage overlap between adjacent grid cells is approximately 0.8 times the main lobe width. The received signal power distribution is calculated by traversing all 1024 grid cells, and the grid cell with the highest power is selected as the optimal beam pointing angle, corresponding to an elevation angle of 75 degrees and an azimuth angle of 120 degrees.

[0069] The current beam pointing angle is 80 degrees elevation and 115 degrees azimuth. The beam adjustment command is calculated as a negative 5-degree elevation adjustment and a positive 5-degree azimuth adjustment. The beam weight adaptive adjustment module, based on a progressive adjustment strategy, corrects 80% of the deviation within the first 25-millisecond coherence time interval, i.e., a 4-degree elevation and 4-degree azimuth adjustment. The adjusted beam pointing angle is 76 degrees elevation and 119 degrees azimuth, with a remaining deviation of 1 degree elevation and 1 degree azimuth from the optimal pointing direction. The remaining deviation is gradually corrected through fine-tuning in subsequent coherence time intervals.

[0070] The beam tracking interval control module determines that it is currently in the second mapping interval based on a channel change gradient of 0.2 dB per millisecond, and sets the tracking interval to 0.65 times the channel coherence time, i.e., 16 milliseconds. After beam alignment is completed, the link quality assessment module monitors the fluctuation range of the received signal strength indication for three consecutive sampling cycles. If the fluctuation range does not exceed ±1 dB, the beam lock is determined to be successful and the system has entered a stable tracking state.

[0071] Under stable tracking conditions, the link bit error rate and throughput are continuously monitored. Assuming the bit error rates for two consecutive statistical periods are 0.0008 and 0.0012, respectively, if the bit error rate in the second statistical period exceeds the threshold of 0.001, the link quality assessment module determines that the link quality has deteriorated and triggers a new round of channel prediction and beam adjustment. This process is repeated cyclically to achieve continuous and stable control of the beam link. Example

[0072] This embodiment describes the specific application of the vehicle-mounted wireless adapter control method of the present invention in a highway scenario, which has different channel characteristics and performance requirements compared with the urban road scenario of Embodiment 1.

[0073] Assume an onboard wireless adapter is installed in a highway service area parking space for a heavy truck, and the terminal device is a wireless access point in a campervan parked in an adjacent parking space. A stable WiFi 6 wireless link needs to be established between the onboard wireless adapter and the terminal device to support high-bandwidth services such as video streaming. The channel environment differs significantly between a stationary and moving vehicle; the channel coherence time is significantly extended to over 100 milliseconds, and multipath effects mainly originate from reflections from service area buildings, while scattering effects caused by vehicle speed are essentially eliminated.

[0074] The parameter configuration of the multi-dimensional feature acquisition and analysis module in static scenarios is adjusted as follows: the sliding window width is set to twice the channel coherence time, i.e., 200 milliseconds, and the window contains 200 sampling points to improve statistical stability. Typical values ​​for the feature vector are: the average value of the received signal strength indicator is -55 dB / mW, the variance is 0.5 squared dB / mW, and the gradient is -0.01 dB / mW per millisecond; the average value of the signal-to-noise ratio is 25 dB, the variance is 0.8 squared dB, and the gradient is -0.02 dB per millisecond; the average value of the channel coherence bandwidth is 80 MHz, the variance is 5 squared MHz, and the gradient is -0.1 MHz per millisecond; and the average value of the channel delay spread is 50 nanoseconds, the variance is 10 squared nanoseconds, and the gradient is 0.5 nanoseconds per millisecond.

[0075] The recurrent neural network model performs prediction calculations based on the feature vectors of the current moment and the previous three historical moments, outputting the predicted channel state value one channel coherence time interval, i.e., 100 milliseconds later. The prediction results show that the channel change is small, and the change in each parameter does not exceed the significant change threshold, thus not triggering the active beam pre-alignment mechanism and maintaining the current beam pointing unchanged.

[0076] The beam tracking interval control module determines that it is currently in the first mapping interval based on the channel change gradient being close to zero, and sets the tracking interval to 100 milliseconds, which is one time the channel coherence time. Compared with the 16-millisecond tracking interval in Example 1, the tracking interval in this example is extended by more than 6 times, significantly reducing system resource consumption and computational overhead.

[0077] Under stable tracking conditions, the bit error rate was consistently maintained at an excellent level below 0.0001, and the throughput remained stable at over 95% of the peak throughput. The link quality assessment module continuously determined that the link was in good condition and did not trigger any adjustment process.

[0078] The above embodiment 2 shows that the vehicle-mounted wireless adapter control method of the present invention can adaptively adapt to different application scenarios. By dynamically adjusting the prediction model parameters and tracking interval strategy, it can achieve low-power lightweight beam management in static scenarios. Example

[0079] This embodiment describes the specific application of the vehicle-mounted wireless adapter control method of the present invention in a multi-user scenario of a vehicle hotspot. Compared with the single-user scenario of Embodiment 1, this scenario requires consideration of the coordinated control of multi-user resource scheduling and beam switching.

[0080] Assuming the in-vehicle wireless adapter acts as a wireless hotspot, simultaneously serving multiple in-vehicle terminal devices, including the driver's mobile phone and tablet, and rear passengers' laptops and gaming devices, beam management in multi-user scenarios requires allocating beam resources among these terminal devices, prioritizing beam quality for devices using high-bandwidth services.

[0081] The multi-dimensional feature acquisition and analysis module is extended to a multi-path parallel processing architecture in multi-user scenarios, with each path corresponding to the channel feature extraction of one terminal device. The parameter configurations for each feature extraction path remain consistent, but the data buffer and statistical calculation modules are independent to avoid data crosstalk between users. The feature vectors are encapsulated with user identification tags before being transmitted to the channel evolution prediction module to support subsequent user-by-user tracking processing.

[0082] The channel evolution prediction module employs a multi-task learning framework in multi-user scenarios. A single recurrent neural network model simultaneously predicts the channel states of multiple terminal devices. Each terminal device shares the feature extraction weights of the hidden layer but has an independent prediction head in the output layer. This multi-task learning framework can improve the model's generalization ability while reducing the overall number of model parameters, making it suitable for deployment in automotive embedded systems.

[0083] In multi-user scenarios, the beam space partitioning and pre-alignment module needs to calculate the optimal beam pointing angles for multiple terminal devices and allocate beam resources according to the service priority and channel quality requirements of each terminal device. The beam scheduling algorithm adopts a weighted fair queue strategy, and the calculation of the weight coefficients comprehensively considers the channel change gradient, service QoS level, and historical throughput contribution of the terminal devices. When the channel of a high-priority terminal device changes significantly, beam resources are scheduled first to complete its beam alignment, while the beam adjustment of low-priority terminal devices is appropriately delayed.

[0084] In multi-user scenarios, the beam weight adaptive adjustment module uses time-division multiplexing to transmit beams to each terminal device in turn. The beam switching period is set to the least common multiple of the tracking intervals of each terminal device to ensure that each terminal device can obtain timely beam updates. When the channel change gradient of a certain terminal device is significantly higher than that of other devices, the beam pointing adjustment frequency of that terminal device is increased accordingly, and the proportion of time it occupies during the beam switching period dynamically increases.

[0085] The beam tracking interval control module and link quality assessment module operate independently in multi-user scenarios, with the tracking status and quality assessment results of each terminal device recorded separately. System-level link quality monitoring calculates the overall system service quality score by aggregating the performance indicators of each terminal device. When the system-level service quality score drops below a threshold, a global beam optimization process is triggered.

[0086] The above embodiment 3 shows that the vehicle-mounted wireless adapter control method of the present invention can be effectively extended to multi-user scenarios, and achieves collaborative beam management of multiple terminal devices through a multi-task learning framework and a multi-user scheduling algorithm.

[0087] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for controlling an in-vehicle wireless adapter based on WiFi 6 technology, characterized in that, Includes the following steps: Multidimensional feature acquisition and analysis of vehicle-mounted channels are performed. Signals from terminal devices are received through the multi-antenna array of the vehicle-mounted wireless adapter. The received signal strength indication value, signal-to-noise ratio value, channel coherence bandwidth value, and channel delay spread value are extracted as basic parameters to describe the channel state. The sliding window mechanism is used to perform time series modeling of the basic parameters. At the end of each window period, the average value, variance, and gradient of the parameters are calculated. These statistics are combined into a multidimensional feature vector that characterizes the short-term changes of the channel. Short-term prediction processing of channel evolution is performed. A prediction model of a recurrent neural network is constructed based on the multidimensional feature vector. The input layer of the recurrent neural network receives the multidimensional feature vectors of the current time and several previous historical time. The output layer outputs the channel state prediction value for several future channel coherence time intervals. When the prediction model determines that the channel will change significantly in the future coherence time interval, an active beam pre-alignment mechanism is triggered. Dynamic beam space partitioning and pre-alignment calculations are performed, and the beam coverage area is discretized into several spatial grid cells. Each grid cell corresponds to a specific beam pointing angle. Based on the signal strength distribution of the future channel state output by the channel prediction module in each grid cell, the optimal beam pointing angle that maximizes the signal power at the receiving end is calculated, and the deviation between the optimal angle and the current beam pointing angle is converted into a beam adjustment command. Active adaptive adjustment of beam weights is performed, and the update amount of beamforming weight vector is calculated according to the beam adjustment command. A progressive weight adjustment strategy is adopted in the beam pre-alignment process. Beam locking and link quality assessment are performed. When the beam pointing adjustment is completed and the received signal strength indicator remains within a stable fluctuation range of a preset threshold for several consecutive sampling periods, the beam locking is determined to be successful and enter a stable tracking state. In the stable tracking state, the bit error rate and throughput of the link are continuously monitored. When the bit error rate exceeds a preset threshold or the throughput decreases by more than a preset amount, the link quality is determined to have deteriorated and a new round of channel prediction and beam adjustment process is triggered.

2. The vehicle-mounted wireless adapter control method according to claim 1, characterized in that, The dimensions of the multidimensional feature vector correspond to the number of the basic parameters. Each basic parameter corresponds to three statistical dimensions, storing its mean, variance, and gradient.

3. The vehicle-mounted wireless adapter control method according to claim 1, characterized in that, The sliding window mechanism dynamically adjusts the window width based on the real-time estimated channel coherence time. When the channel coherence time is shortened, the window width is reduced accordingly to improve the temporal resolution of feature acquisition. When the channel coherence time is extended, the window width is expanded accordingly to improve the statistical stability of feature data.

4. The vehicle-mounted wireless adapter control method according to claim 1, characterized in that, The input layer of the recurrent neural network receives multidimensional feature vectors from the current time and several previous historical time points, and the output layer outputs channel state prediction values ​​for several future channel coherence time intervals. The predicted output data is used to guide the pre-adjustment of beam pointing.

5. The vehicle-mounted wireless adapter control method according to claim 1 or 4, characterized in that, The recurrent neural network adopts a three-layer topology design for its hidden layers. The three hidden layers are fully connected, and the output of all neurons in the previous layer serves as the input of all neurons in the next layer.

6. The vehicle-mounted wireless adapter control method according to claim 1, characterized in that, The angular spacing of the spatial grid cells is set to a certain proportional multiple of the beam main lobe width, and there is an effective beam coverage overlap area between adjacent grid cells.

7. The vehicle-mounted wireless adapter control method according to claim 1 or 6, characterized in that, The beam adjustment command includes pitch angle adjustment in the horizontal dimension and azimuth angle adjustment in the vertical dimension.

8. The vehicle-mounted wireless adapter control method according to claim 1, characterized in that, The beamforming weight vector controls the signal transmission phase and amplitude of each antenna element in the multi-antenna array to form a beam pointing in a specific direction. The update of the weight vector follows constraints to ensure that the total transmit power of the array antenna remains constant.

9. The vehicle-mounted wireless adapter control method according to claim 1 or 8, characterized in that, The initial adjustment step size of the progressive weight adjustment strategy is adaptively determined based on the predicted channel change amplitude. When the predicted change amplitude is large, the initial adjustment step size is increased to speed up beam alignment, and when the predicted change amplitude is small, the initial adjustment step size is decreased to improve the stability of beam adjustment.

10. The vehicle-mounted wireless adapter control method according to claim 1, characterized in that, It also includes the step of adaptive optimization control of beam tracking interval, which dynamically adjusts the beam tracking trigger interval according to the channel change gradient in the multidimensional feature vector. When the channel change gradient is low, the tracking interval is increased to reduce unnecessary beam adjustment overhead, and when the channel change gradient is high, the tracking interval is decreased to improve the response speed of beam tracking.

Images